Control strategies for order–disorder phase transition in crowd evacuation

Crowd evacuation can undergo abrupt, hazardous shifts between ordered and disordered motion, yet how network topology and targeted interventions jointly shape this transition remains unclear. We integrate a dynamic, weighted small-world contagion layer with an extended social force model and design adaptive, topology-aware interventions that target high-degree (HD) and high $k$-shell (HK) nodes. Simulations map a risk-induced transition: low risk permits spontaneous recovery, medium risk triggers a critical collapse of alignment, and high risk locks the system into disorder. Targeting a small fraction of agents (10%–20%) lowers collective impatience and preserves alignment, with HD retaining more benefit at higher densities. In dual-exit rooms under high risk — where a short random-walk disorientation rule and a pressure–emotion coupling are active — raising alignment via targeted control reduces evacuation time, re-balances exit usage, and lowers peak contact forces. These effects remain robust across exit-width changes, increases in the long-range contagion probability $p\left(t\right)$, and moderate inter-individual heterogeneity. Analyses of real crowd recordings further show that HD and HK selections overlap only partially, supporting dynamic, hybrid policies that cover both dense cores and structural bridges. Together, the results provide a topology-aware control framework that links network structure to emergent evacuation behavior, with direct implications for planning, public safety, and crowd resilience.